🚀 go-pugleaf

[I took this off comp.sys.super. Steve]

  [Dr. David Kahaner is a numerical analyst on sabbatical to the 
   Office of Naval Research-Asia (ONR Asia) in Tokyo from NIST.  The 
   following is the professional opinion of David Kahaner and in no 
   way has the blessing of the US Government or any agency of it.  All 
   information is dated and of limited life time.  This disclaimer should 
   be noted on ANY attribution.]

  [Copies of previous reports written by Kahaner can be obtained using
   anonymous FTP from host cs.arizona.edu, directory japan/kahaner.reports.]

To: Distribution
From: 
 David K. Kahaner
 US Office of Naval Research Asia
 (From outside US):  23-17, 7-chome, Roppongi, Minato-ku, Tokyo 106 Japan
 (From within  US):  Unit 45002, APO AP 96337-0007
  Tel: +81 3 3401-8924, Fax: +81 3 3403-9670
  Email: kahaner@cs.titech.ac.jp
Re: Supercomputer items from Japan, Fujitsu and ICFD.
22 Sept 1992
This file is named "jsuper.992"

ABSTRACT. Brief update on Fujitsu's new supercomputer (VPP) and Tokyo's
Institute for Computational Fluid Dynamics.

FUJITSU'S NEW SUPERCOMPUTER

Fujitsu has announced their latest high performance computer. This
appears to be a cluster/crossbar system using a modest number of
extremely powerful individual processors. Each processor includes a
vector processor and has peak performance of 1.6GFLOPs and 256MBytes
memory.  Processors use both BiCMOS LSI chips (72,000 gates and 200
picosecond delay time) as well as 45 GaAs LSI chips (25,000 gates and 60
picosecond delay time).  One board contains both vector and scalar units
as well as memory for compactness.

Groups of 7 or 8 processors can be combined (the actual number depends
on whether one counts the control processors), up to 222 units with a
peak performance of over 355GFLOPs. All processors are designed as LIW
(Long Instruction Word) RISC units, with asynchronous operation of
scalar, vector and floating point units as well as communication. The
system is networked by a crossbar, and communication units within each
processor have a peak performance of 800MBytes/second.

Fujitsu has named this series VPP500 and claimed that it can be
purchased as a back-end to Fujitsu's current VP2000 line for current
Fujitsu users. Users can compile on the VP2000 and use it as a file
server to the VPP. Fujitsu will provide Unix System Vr4 as an operating
system and claims that software written for their VP systems can also be
used on the VPP. The machine will use fully IEEE standard arithmetic
(although the front end VP200 might stay with IBM format).  Delivery is
announced as September 1993, and Fujitsu hopes to sell 50 systems within
three years (20 in Japan and 30 exported). Fujitsu will OEM these to
Siemens Nixdorf Information Systems of Germany. Rental prices are
announced as $244,000--$4.1M per month, with a purchase price of about
$125M US for the largest system.

In an earlier report ["sacad.92", 1 July 1992], I discussed several
aspects of a Numerical Wind Tunnel (NWT) computer to be developed for
the Japanese National Aerospace Laboratory, and that NAL was working
with Fujitsu in the development of such a machine. This appears to be
the new VPP. This month in Paris at the Fujitsu International
Supercomputer Users Group Meeting, a description of NAL's NWT showed
that it will have the same compiler directives as the VPP; in Feb 1993
NAL hopes to get a system with 140 processors and is expecting 64GFLOPs
sustained for their real applications.

At the Paris meeting the ECMWF (European weather prediction community)
discussed their needs. They are certainly interested in a VPP to replace
their C90 system (undergoing acceptance tests) in about two years.
Fujitsu would be happy to sell them one and is likely to concentrate on
marketing outside of the US. Of course, by then there will be other
machines around too.

The development of the VPP follows a trend that we predicted in our
assessment report on Japanese supercomputers [see "jhpc-sc.92", 28 June
1992].

(1) Focus on exploiting excellent device and packaging technology to
push individual processor performance.

(2) A conservative trend with respect to very large numbers of
processors, which are still viewed as difficult to program and use. The
VPP system is really on a different track from Fujitsu's AP system; the
latter can have up to 1K processors in a mesh architecture.

(3) Use of some GaAs in new generation machines, but conservatively.


INSTITUTE OF COMPUTATIONAL FLUID DYNAMICS (ICFD)

Dr. Kunio Kuwahara (sometimes Romanized as Kuwabara) has made headlines
over the past few years. Using land in Haramachi that he owns (in
central Tokyo), Kuwahara built (next to his house) a center for
supercomputer applications in fluid dynamics. Leasing first one then
another and eventually seven supercomputers (estimates are about $80M US)
he used the land as collateral for bank loans. In addition to his own
research he leased out excess cycles, mostly to automobile, steel, and
chemical makers. Readers of my reports may remember that Dr. Hideo
Yoshihara, who was in this office for two years ending in 1990,
performed many benchmarks of Japanese supercomputers collaborating with
scientists at ICFD. 

Alas, a combination of hard economic times, falling land values, and
direct supercomputer purchases by some of his customers eventually
caused ICFD's money to run out. Now all six of the installed leased
machines have been removed, leaving only a Thinking Machines CM-2, which
was purchased outright. The official reason stated is that ICFD needed
better management because of financial difficulties. Current plans are
to give up the time-rental business and focus on consulting work using
parallel computers, such as Thinking Machines CM-2 and others. ICFD will
also devote some of its business activities to selling workstation
compatible software. 

Kuwahara is still a flamboyant and colorful figure, once commenting that
he didn't need any fancy software or techniques, just more cycles. He
asserted that US supercomputer makers such as Cray Research are "small"
companies, building products largely by hand, and claimed that their
machines were expensive and temperamental, whereas Japanese companies
with large product lines have excellent production technology, in-house
semiconductor technology, and can spread development costs over an
entire product line.

---------------------------END OF REPORT------------------------------